How do servers with redundant power supplies balance consumption?

Question

I have several servers from the HP DL360 line (generations 5-8). Each of these servers has two power supplies installed. The 2 power supplies in each server are fed from different circuits.

My question is will the power draw be roughly balanced between these two circuits, or do the servers consider one power supply to be a "primary" and the other to be a "backup" with a smaller power draw?

Remember that, in practise, you will have to design both of your circuits strong enough to handle the full load in case the other circuit goes down, or you'll miss the whole point of redundant power supplies. So i don't really understand if there's any decision that depends on the answer to this. — Guntram Blohm, Jan 14 '15 at 19:20
@GuntramBlohm There are situations where power is constrained in a co-location facility or cases where one needs to manually balance the power distribution across PDU feeds. The servers the OP is using allow you to pin a particular PSU as primary through the BIOS. — ewwhite, Jan 14 '15 at 19:38
Also, sometimes power supply redundancy is more about keeping the attached device on, versus the threat of facility power loss. That's another situation where design for fully redundant load levels may not be necessary. — ewwhite, Jan 15 '15 at 00:54

score 50 · Accepted Answer · edited Apr 13 '17 at 12:14

I'm going to give an HP Proliant-specific answer here, since the OP is asking about the HP product line.

Let's use the example of an HP DL360p Gen8 (also applies to G6, G7 and Gen9 servers):

You have the option to configure the Redundant Power Supply Mode in the HP Rom-Based Setup Utility (press F9 at boot) with:

Balanced Mode
High Efficiency Mode

Of the two main options, there's some granularity in the High-Efficiency Mode:

High Efficiency Mode provides the most power efficient operation with redundant power supplies by keeping one power supply in standby mode at lower power usage levels.
Balanced Mode shares the power equally between both power supplies.
In addition, the "Auto" mode chooses between one PSU or the other as primary, depending on the serial number. It's a way to randomize the distribution in a datacenter situation with multiple servers.
Also see: Maxing out both PDUs in a rack with redundant power

An example of the Balanced Mode on a busy server:

hpasmcli> SHOW POWERSUPPLY
Power supply #1
    Present  : Yes
    Redundant: Yes
    Condition: Ok
    Hotplug  : Supported
    Power    : 110 Watts
Power supply #2
    Present  : Yes
    Redundant: Yes
    Condition: Ok
    Hotplug  : Supported
    Power    : 95 Watts

An example of the High Efficiency Mode on an idle server:

hpasmcli> SHOW POWERSUPPLY
Power supply #1
    Present  : Yes
    Redundant: Yes
    Condition: Ok
    Hotplug  : Supported
    Power    : 50 Watts
Power supply #2
    Present  : Yes
    Redundant: Yes
    Condition: Ok
    Hotplug  : Supported
    Power    : 20 Watts

An example of the High Efficiency Mode on a busy server:

hpasmcli> SHOW POWERSUPPLY
Power supply #1
    Present  : Yes
    Redundant: Yes
    Condition: Ok
    Hotplug  : Supported
    Power    : 90 Watts
Power supply #2
    Present  : Yes
    Redundant: Yes
    Condition: Ok
    Hotplug  : Supported
    Power    : 20 Watts

Detailing the relative efficiency of single PSU, load-balanced PSUs and High Efficiency mode on a 750W power supply.

Why would you ever **not** use High Efficiency mode? What advantages does balanced give? — Oli, Jan 15 '15 at 12:12
@Oli The readings are from two different workloads. The **Balanced Mode** output was from an incredibly busy system loaded with PCIe cards. The **High Efficiency** readings were to illustrate that one PSU is kept at a nominal value regardless of workload. — ewwhite, Jan 15 '15 at 12:14
Fair enough, but given that most PSUs operate most efficiently nearest their rated loads, why would you ever want balanced? Running at <50% is going to nuke efficiency and I've never seen anything to suggest PSUs' wear-rates depend on their load. They're going to wear near the same rate whatever. I can't see a good reason for it in a redundancy setup. (Sorry if that sounds accusatory; this is really just a curiosity, it's a good answer without it). — Oli, Jan 15 '15 at 12:18
The PSUs in the examples are 750W. They never get anywhere close to their rated load figures. You're right about the choice. Balanced is the default, so I'd assume that's why people use it. I tend to prefer high-efficiency mode, though, because it's... well, [*more efficient*](http://h30507.www3.hp.com/t5/Reality-Check-Server-Insights/More-for-Your-IT-Dollar-High-Efficiency-Mode/ba-p/80739#.VLexb8bDvSx). — ewwhite, Jan 15 '15 at 12:26
@Oli, in my experience peak efficiency is usually around 50% of rated load, so balancing is good if you're consistently near 100% of a single PSU's capacity. High-efficiency mode is essentially active/passive failover, which is a win if you're closer to 50% of a single PSU's capacity than 100%. The curve above for dual 750W bears this out. — Jeffrey Hantin, Jan 15 '15 at 23:55

score 12 · Answer 2 · answered Jan 14 '15 at 14:47

it depends

If you go into setup (F10) you can choose the power management mode and usually there's a few options including the active/passive options you suggest as well as a balanced mode. I buy blades and you do this at the blade-enclosure level and there seems to be more options but essentials it depends :)

score 5 · Answer 3 · answered Jan 14 '15 at 17:47

Every server is different. Some automatically balance between the two and some use the primary PSU exclusively until it fails and then switches to the secondary PSU. In most cases there is a bios option to configure how the server treats the PSU's. If for some reason you don't have that option you can find out by putting a clamp ammeter on each power cable and measuring the current draw.

How do servers with redundant power supplies balance consumption?

3 Answers3

Linked